Cooperative Steering Control System for Robust Obstacle Avoidance in Cluttered Off-Road Environments

Abstract

Emerging intelligent vehicle technologies make sig-nificant contributions to alleviating driving workloads and im-proving the road safety. At current stage of development, driver-vehicle collaboration related to the semi-autonomous driving concept becomes a reliable and popular approach. Here we propose a cooperative steering control scheme to address the challenging human-machine interaction problem in vehicle control design, especially for lateral collision avoidance tasks. Instead of relying on the conventional planning-tracking paradigm, our method dynamically balances the control authority between the human driver and the cooperative agent according to real-time evaluation on collision risks. Specifically, a novel obstacle avoidance algorithm named as steering field histogram (SFH) is designed to incorporate nonlinear vehicle lateral dynamics. In order to better understand driver's steering intentions, a Gated Recurrent Unit (GRU) based learning framework is deployed to capture the temporal-dependent features in historical driving data. Eventually, results of driver-in-the-loop experiments show that the shared control system can decrease the collision rate by 96.1% and the driver's workload by 50.3%. Moreover, up to 79.9% of the time remains to be free for drivers to take full control of the vehicle and to employ their own decision-making behaviors.